JP5321939B2 - Image quality inspection apparatus and image quality inspection method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image quality inspecting device capable of properly inspecting the image quality of a two-dimensional image. <P>SOLUTION: Image data outputted from a CCD camera 1 is inputted to the image quality inspecting device 2. The image quality inspecting device 2 is equipped with a correction means 21 for performing correction so as to suppress the effect of discontinuity in the boundary of an inspection region, by using a window function with respect to the inspection region of the two-dimensional image becoming an inspection target; a flaw intensity calculation means 22 for calculating flaw intensity by integrating the frequency characteristics of sight sensitivity with the frequency component of the two-dimensional image becoming the inspection target, with respect to the image after being corrected by the correction means 21, and a determination means 23 for determining the image quality of the two-dimensional image, on the basis of the flaw intensity calculated by the flaw intensity calculation means 22. <P>COPYRIGHT: (C)2009,JPO&amp;INPIT

Description

  The present invention relates to an image quality inspection apparatus and an image quality inspection method for inspecting the image quality of a two-dimensional image.

  In the manufacturing process of the display device, the defect inspection is performed on the panel alone before moving to the assembly process. This is because pixel defects generated in the panel manufacturing process are inspected, and defective parts are repaired before assembly, or a panel that cannot be repaired is not sent to the next process but discarded.

  In conventional visual inspections by inspectors, there are many individual differences and variations, and it is difficult to inspect with high accuracy. In addition, with the recent increase in size and definition of display devices, a wide range can be quickly achieved. Since it is also difficult to inspect, an alternative to an automatic inspection device is also desired in the inspection of mura defects.

  The inspection target is a mura defect that occurs on a display device, an imaging device, or a projection device. Unevenness is caused by various factors such as a change in the thickness of the glass substrate in the manufacturing process, a change in the thickness of the color filter, and an alignment error of the liquid crystal. There are various shapes, orientations, occurrence positions, and sizes (frequencies) such as lump-like unevenness and strip-like unevenness.

  2. Description of the Related Art As a conventional automatic inspection apparatus that uses a display device as an inspection target, an apparatus that captures an inspection target device by combining a photographic lens with a high-pixel monochrome CCD camera at a position facing the inspection target device is known. The camera is installed in a state where the tilt is adjusted with respect to the panel to be inspected, both vertically and horizontally.

  For the captured image, “image correction processing” for performing noise correction of image data, “defect candidate extraction processing” for extracting defect candidates, and calculating feature quantities such as area, volume, and contrast of the extracted defect candidates “ It is output as inspection result data through “feature amount calculation processing” and “determination processing” in which the calculated feature amount is compared with a preset threshold value to determine whether or not it is a defect.

  FIG. 17 is a flowchart showing an inspection procedure by the image quality inspection apparatus using the MTF characteristic with respect to luminance.

  This method expresses the strength of a defect as a ratio between the defect contrast and the recognition limit (JND) contrast for the defect shape, and is expressed as an objective measurement value in the unit of “Sem”. Yes, considering various changes (background brightness, defect shape, etc.) of inspection object.

  The unit of measure “Semu” was proposed in the SEMI standard in 2002.

  First, the MTF characteristic is a sensitivity characteristic represented by a reciprocal of a limit contrast change that can be recognized and sensed by a human in a bright and dark sinusoidal fringe pattern as shown in FIG. The human visual sensitivity characteristic with respect to is a characteristic curve having a peak at a certain frequency as shown in FIG. That is, it is shown that human vision is sensitive to certain details.

  Specifically, as shown in FIG. 18C, when a sine wave having a wave number within a certain viewing angle is displayed, the spatial frequency is

Spatial frequency (cpd) = wave number / viewing angle (degrees) (1)
In addition, the recognition limit (C _JND , JND: just noticeable difference) contrast is defined as “ΔL” as the minimum detectable luminance change and “L _BG ” as the background luminance.

C _JND = ΔL / L _BG × 100 (2)
It is represented by The relationship between the reciprocals of the equations (1) and (2) is the MTF characteristic “S _MTF ”.

  However, in order to use it for image quality inspection, it is necessary to expand to a two-dimensional MTF characteristic. Regarding the two-dimensional MTF characteristics, a difference in sensitivity depending on the angle has been reported.

  In the “image correction process” of FIG. 17, in step S001, the input image data is subjected to reduction processing and smoothing processing for optical characteristic correction, shading correction, correspondence to defect size, high frequency noise removal, and the like. Etc. are pre-processed.

  In the “defect candidate extraction process”, in step S002, the luminance difference of the defect candidate is enhanced by an enhancement filter such as a Laplacian type, a Sobel type, or a projection. Next, in step S003, the image in which the luminance difference is enhanced is binarized with a threshold value, and a portion that exceeds the threshold value (or a portion that does not satisfy the threshold value) is extracted as a defect candidate and output.

  In the feature amount calculation process, a defect shape obtained by normalizing the defect candidates extracted in the “defect candidate extraction process” is generated in step S004. Normalization is performed, for example, from the maximum brightness inside the defect candidate and the brightness around the defect candidate.

  In step S005, the normalized defect shape is subjected to processing such as two-dimensional Fourier transform to extract a frequency component “Power (u (h, v))” that forms the defect shape. Here, u (h, v) is a spatial frequency.

On the other hand, in step S006, the spatial frequency “(u (h, v))”, background luminance “L _BG ”, defect size “(X, Y)”, angular direction “φ”, and the like are used as parameters. MTF characteristics “S _MTF (u (h, v), L _BG , X, Y, φ)” expanded to dimensions
Calculate

In step S 007, the defect contrast “C _X ” is calculated from the luminance value of the defect area and the luminance value of the periphery of the defect from the equation (2).

Next, in step S008, the obtained frequency component “Power (u (h, v))” and the two-dimensional MTF characteristic “S _MTF (u (h, v), L _BG , X, Y, φ)” are displayed.
From this, the recognition limit contrast “C _JND ” for the defect shape is calculated by the following equation.

Finally, in step S009, the defect intensity (Sem) is calculated from the defect contrast “C _X ” and the recognition limit contrast “C _JND ” for the defect shape by the following equation.
Semu = C _X / C _JND Formula (4)

  In the “determination process”, in step S010, the defect strength (Semu) calculated in the “feature amount calculation procedure” is compared with a preset threshold value to determine whether or not the defect is present. If it is determined that there is, information on the detected defects is collected into output data in step S011.

The “feature amount calculation process” and the “determination process” are executed for all candidate areas extracted in the “defect candidate extraction process”.
JP 2005-326323 A Journal of Sciology Vol.60, No.2, p129-136, 2006 Quantification method of intensity of luminance unevenness based on visual system spatial frequency characteristics, Takuto Kishi SEMI D31-1102 Definition of measurement unit (Semu) of luminance unevenness in FPD image quality inspection Journal of the Japan Society of Photography, Vol.65, No.2, p121-127, 2002 Directional dependence of visual system spatial frequency response (1), Tetsuya Ishihara Journal of the Japan Photography Society Vol.65, No.2, p128-133, 2002 Directional dependence of the spatial frequency response of the visual system (2), Tetsuya Ishihara

However, the above inspection method has the following problems.
(1) Consider, for example, an image in which a white defect exists in the center and black vertical stripe defects exist on the entire surface. In this way, when there is a mixture of blocky defects and periodic defects, it is not possible to determine what should be the brightness of the defect and the brightness of the surroundings, and it is not possible to calculate an appropriate defect contrast. The SEMU value cannot be calculated.
(2) In the case of an image of a substantially uniform vertical stripe defect having a certain period and a certain contrast, the same Semu value should be calculated if the candidate area size for which the intensity calculation is originally performed is the same. Even if the candidate region size to be performed is the same, the calculated Semu value differs depending on the cutout region.
(3) When an actual inspection object is inspected, since an imaging system is included, rough noise is included due to camera noise or flickering or variation of the object itself. In such a case, there is a problem that the Semu value cannot be accurately calculated due to the influence of coarse noise.

  An object of the present invention is to provide an image quality inspection apparatus capable of appropriately inspecting the image quality of a two-dimensional image.

Image quality inspection apparatus of the present invention, the image quality inspection apparatus for inspecting the quality of the two-dimensional image, the inspection area of the two-dimensional image to be inspected, the discontinuity at the boundary of the inspection area by using a window function A correction unit that performs correction to suppress the influence, a determination unit that determines an image quality of the two-dimensional image based on the image corrected by the correction unit, and a plurality of two-dimensional images that overlap each other. Control means for controlling to repeat the correction by the correction means and the determination by the determination means while sequentially switching the inspection areas divided into regions, the frequency characteristics of visual sensitivity, and the correction means are corrected. and a defect intensity calculating means for calculating the defect intensity by multiplying the frequency components of the image after the two-dimensional image Ri inverse colored image der, before Symbol judging means, before Determining the image quality of the two-dimensional image on the basis of the defect intensity calculated by the defect intensity calculating means, before Symbol defect intensity calculating means, the defect intensity of only the dominant frequency component in the two-dimensional frequency region of the examination region calculated that you characterized that.
According to this image quality inspection apparatus, since the influence of discontinuity at the boundary of the inspection area is suppressed using the window function, the image quality can be appropriately evaluated without being affected by the inspection area.
Further , according to this image quality inspection apparatus, the defect intensity is calculated by integrating the frequency characteristic of the visual sensitivity and the frequency component of the two-dimensional image to be inspected, so that it is objective and easy to understand intuitively. Defect strength can be obtained as a measured value.

The dominant frequency component may be extracted by multiplying a Fourier transform image of the inspection region and a binarized image of the Fourier transform image.

  The defect intensity calculating means calculates the first defect intensity by integrating the frequency characteristic of the visual sensitivity and the frequency component of the first two-dimensional image to be inspected, and the frequency characteristic of the visual sensitivity. And the second defect intensity is calculated by integrating the frequency components of the second two-dimensional image to be inspected, and the determination means calculates the first defect intensity calculated by the defect intensity calculation means. The image difference between the first two-dimensional image and the second two-dimensional image may be evaluated by comparing the second defect strength.

  The two-dimensional image may be a luminance image.

The size of the inspection area may be variable.

  According to the image quality inspection apparatus of the present invention, since the influence of discontinuity at the boundary of the inspection area is suppressed using the window function, the image quality can be appropriately evaluated without being affected by the inspection area.

  According to the image quality inspection apparatus of the present invention, since the defect strength is calculated by integrating the frequency characteristics of the visual sensitivity and the frequency component of the two-dimensional image to be inspected, it is objective and easy to understand intuitively. Defect strength can be obtained as a measured value.

  Hereinafter, an embodiment of an image quality inspection apparatus according to the present invention will be described with reference to FIGS.

  FIG. 1 is a block diagram illustrating a configuration of an image quality inspection apparatus according to an embodiment.

  A high-pixel monochrome CCD camera 1 combined with a photographic lens is disposed at a position directly opposite the inspection target device 3, thereby imaging the inspection target device. The CCD camera 1 is installed in a state in which the tilt is adjusted accurately and vertically with respect to the panel to be inspected.

  Image data output from the CCD camera 1 is input to the image quality inspection apparatus 2.

  The image quality inspection apparatus 2 includes a correction unit 21 that performs correction so as to suppress the influence of discontinuity at the boundary of the inspection region using a window function with respect to the inspection region of the two-dimensional image to be inspected, Defect strength calculating means 22 for calculating the defect strength by integrating the frequency characteristic of visual sensitivity and the frequency component of the two-dimensional image to be inspected with respect to the image corrected by 21, and defect strength calculating means And determining means 23 for determining the image quality of the two-dimensional image based on the defect intensity calculated by 22.

  The function of the correcting means 21 is the process of steps S104 to S105 described later, the function of the defect strength calculating means 22 is the process of steps S106 to S110 described later, and the function of the determining means 23 is the “determination process” described later. Each corresponds.

  FIG. 2 is a flowchart showing an inspection procedure by the image quality inspection apparatus using the MTF characteristic with respect to luminance.

  In “image correction processing”, preprocessing such as optical characteristic correction and shading correction is performed on the input image data in step S101.

  In the “image correction process”, an image may be corrected using a camera output value / brightness conversion parameter. For example, the camera output value can be converted into actual luminance using the camera output value / luminance conversion parameter stored in the storage device. In this case, the image quality inspection with high accuracy can be performed by executing all or part of the “image correction processing”, “feature amount calculation processing”, and “determination processing” using the actual luminance value information.

  Next, in the “defect candidate extraction process”, in step S102, the brightness difference of the defect candidate is enhanced by an enhancement filter such as a Laplacian type, a Sobel type, or a projection. Next, in step S103, the image in which the luminance difference is emphasized is binarized with a threshold, and a portion exceeding (or not satisfying) the threshold is extracted as a defect candidate and output.

  The image quality inspection apparatus of the present invention is characterized by “feature amount calculation processing” described below.

  In the “feature amount calculation process”, in step S104, a contrast image with an average luminance of the calculation area is generated for the calculation area including the periphery of the defect candidate according to Expression (5).

Here, imgLumi (x, y) is a luminance conversion image of the calculation area converted to cd / m ² units, and aveLumi is the average luminance of the calculation area.

  Next, in step S105, the contrast image is corrected by a window function in order to cope with signal discontinuity during FFT (Fast Fourier Transform). For example, a two-dimensional Akaike window image is generated in accordance with the calculation area, and is multiplied by the contrast image.

imgCorr (x, y)
= ImgCont (x, y) · imgWin (x, y) (6)

  Here, imgWin (x, y) is a two-dimensional Akaike window image.

  FIG. 3 shows the behavior of window function correction. FIG. 3A shows an original image, and FIG. 3B shows a corrected image by a window function. As shown in FIG. 3A, in the original image 25A, luminance discontinuity is observed at the outer peripheral edge of the calculation area 24, whereas in the corrected image 25B by the window function, at the outer peripheral edge of the calculation area 24. The effect of discontinuity is reduced. Thereby, it is suppressed that the calculation result of defect strength changes with cutout areas.

Next, in step S106, processing such as two-dimensional Fourier transform is performed on the corrected contrast image to extract a frequency component “Power (u _h , u _v )” that forms the contrast image. Here, “(u _h , u _v )” is a spatial frequency.

Next, in step S107, binarization processing is performed on the FFT image “Power (u _h , u _v ))” in order to extract dominant frequency components in the calculation region. For example, binarization is performed as follows using the average value and standard deviation of the FFT image.
imgBinFET (u _h , u _v )
= Power (u _h , u _v ) ≧ AveFFT + StdevFFT × α (7)

  Here, AveFFT: average value of FFT image, StdevFFT: standard deviation of FFT image.

  Next, in step S108, the dominant frequency component in the calculation region is extracted by multiplying the FFT binary image and the FFT image. The purpose is to remove excessive high frequency components caused by rough noise or the like while leaving strong harmonic components necessary to form the defect shape.

Power (u _h , u _v )
= Power (u _h , u _v ) · imgBinFFT (u _h , u _v ) (8)

On the other hand, in step S109, the MTF characteristics “S _MTF (u _h) expanded two-dimensionally using the spatial frequencies u _h , u _v , background luminance L _BG , defect size X, Y, angle direction φ, and the like as parameters, respectively. , U _V , L _BG , X, Y, φ) ”
Calculate

  Whereas the MTF positioning in the conventional method is an amplitude conversion function (MTF: Modulation Transfer Function) that converts a normalized defect shape with an amplitude of 1 to a JND contrast, The positioning of the MTF in the invention becomes the human contrast sensitivity (reciprocal of JND contrast) itself with respect to the frequency component, and can be said to be an amplitude conversion function that converts the contrast amplitude of each frequency component in the FFT image into a numerical value that is a multiple of JND.

Next, in step S110, from the obtained two-dimensional MTF characteristics S _MTF (U _h , U _V , L _BG , X, Y, φ) and the frequency component Power (u _h , u _v ), the defect strength is calculated by the following equation. (Semu) is calculated.

  Next, in the “determination process”, in step S111, the defect intensity (Semu) calculated in the “feature amount calculation process” is compared with a preset threshold value to determine whether or not the defect is present. If it is determined that the defect is a defect, the information of the detected defect is collected into output data in step S112.

  The “feature amount calculation process” and the “determination process” are executed for all candidate areas extracted in the “defect candidate extraction process”.

  As described above, according to the image quality inspection apparatus of the present embodiment, it is not necessary to calculate the defect contrast, and it is not necessary to strictly identify the defect area and the peripheral area. Therefore, it is possible to realize a high-accuracy image quality inspection apparatus that quantifies the defect intensity as an objective measurement value that is easy to understand intuitively by the user.

  Further, by performing correction using the window function, it is possible to provide an image quality inspection apparatus capable of calculating a stable defect intensity for a defect having the same intensity even when the cutout region is changed.

  Furthermore, by correcting the frequency components, excessive high frequency components caused by rough noise and the like are removed while leaving strong harmonic components necessary to form the defect shape. It is possible to provide an image quality inspection apparatus capable of reducing the influence of noise and the like and calculating a stable defect strength for a defect having the same strength.

  Next, FIG. 4 is a diagram showing an image quality inspection method in which inspection is performed while sequentially shifting the inspection areas for extracting defects so as to overlap, and FIG. 5 is a flowchart showing this inspection procedure. In FIG. 5, the same step numbers are assigned to the processes similar to those in FIG.

  In the inspection procedure shown in FIG. 5, an “area setting process” is employed instead of the “defect candidate extraction process” in FIG.

In the “area setting process”, in step S202, the size, movement width, and the like of the divided areas that are the targets of the “feature amount calculation process” and “determination process” are set. Here, as shown in FIG. 4A, it is divided into a plurality of overlapping inspection areas 3a, 3a,..., And the target inspection area is moved for all panel areas as shown in FIG. The operation is repeated. For example, as shown in FIGS. 4 (a) and 4 (b), when the inspection area 3a is set so as to be shifted by half a block, a defect exists across the inspection areas 3a, 3a,. Also correspond. At this time, if the size of the inspection region 3a is increased, it is possible to deal with a large-size defect or a low-frequency defect, and if the size is reduced, it is possible to deal with a small-size defect or a high-frequency defect.

  In this way, by setting the inspection areas 3a, 3a,... So as to overlap with each other, the area that the person feels conspicuous without being caught by the defect shape / size is extracted as the defect area, and the user is intuitive It is possible to quantify the defect strength as an objective measurement that is easy to understand. In addition, since the defect strength is determined using the frequency components of the entire divided area, not only single defects that do not have a specific frequency component, but also periodic defects, without being caught by the frequency, direction, Detection is possible.

  In addition, it is possible to detect a new defect because it is determined whether or not it is a defect according to the conspicuousness of each divided area without being aware of the defect shape, reducing the development cost of a new inspection algorithm, and the inspection algorithm development period. Can be shortened. In addition, since there is no extraction process that depends on the defect shape and size, it is possible to achieve a significant reduction in the number of tuning steps related to defect area extraction.

  As shown in FIG. 1, the recognition limit sensitivity model parameter is stored as an external file 29, and the recognition limit sensitivity model parameter is input from the external file 29 to the defect strength calculation means 22 and the determination means 23. Here, the recognition limit sensitivity model parameter is passed to some or all of defect candidate extraction processing, feature amount calculation processing, and determination processing.

  Thus, by inputting the recognition limit sensitivity model parameter from the outside, the maintainability is improved and the sensitivity calculation with high accuracy becomes possible. For example, by measuring the characteristics of a device to be inspected in real time and generating and using a recognition limit sensitivity model parameter from the data, it is possible to calculate the defect intensity corresponding to variations in characteristics between devices and changes over time.

  FIG. 6 is a flowchart showing an inspection procedure when a multi-step smoothing reduction function is added to the procedures shown in FIGS. 4 and 5. In FIG. 6, processes corresponding to the steps in FIG.

  In the “image correction function”, optical property correction and shading correction are performed on the image data input from the “image capturing means” as preprocessing in step S301, and then in step S302, the defect size / Reduction processing, smoothing processing, and the like are performed in order to cope with frequency and remove high-frequency noise.

  The subsequent processing (steps S104 to S112) is repeatedly executed according to the number of smoothing / reducing processes executed.

  In this way, multi-stage smoothing reduction is performed sequentially, and subsequent “region setting processing”, “feature amount calculation processing”, and “judgment processing” are repeatedly executed for each reduced image, thereby allowing a wide range of size and frequency defects. Is possible.

  FIG. 7 is a block diagram illustrating a configuration of an image quality inspection apparatus that inspects color unevenness. Components similar to those in FIG. 1 are denoted by the same reference numerals.

  A turret 11 having RGB color filters 11R, 11G, and 11B is inserted between the photographing lens of the monochrome CCD camera 1 and the inspection target device 3, and the turret 11 is supported by a rotation driving unit such as a stepping motor or a servo motor. It has a structure that can be rotated and positioned at high speed. The turret rotates in a programmable manner in conjunction with the shooting sequence, allowing arbitrary filter selection at high speed.

  Instead of the configuration in which the turret and the monochrome camera are combined as shown in FIG. 7, a configuration with a 3CCD camera having a built-in color filter without using the turret is also possible.

  Also, instead of the RGB color filter, consider the spectral characteristics of a monochrome camera and use an X, Y, Z filter designed to directly capture XYZ tristimulus images, or use a two-dimensional colorimeter. For example, it is possible to directly input an image of XYZ tristimulus values.

  As shown in FIG. 7, the image quality inspection apparatus 2 </ b> A to which an RGB color image is input includes an image correction unit 21 that generates an image in which lens aberration is corrected with respect to each input image data, and is corrected. An opposite color image generating means 26 for converting the image into an opposite color component image and a feature amount calculation area are set, and a defect ratio for the opposite color component is set as a contrast ratio for the recognition limit contrast of each opposite color component for each divided area. Defect intensity calculation means 22 for calculating the intensity, and the total defect intensity as color unevenness is calculated from the calculated opposite color defect intensity and compared with a preset threshold value to determine whether or not the defect is present. Determination means 23 is provided.

  As described above, the image quality inspection apparatus 2A is provided with the opposite color image generation means 26, which converts the three input images into three opposite color component images, and the defect intensity calculation means for each image. 22 is operating.

  Here, the opposite color component is Herring's opposite color theory (in the retina, there are three types of receivers that respond to white-black, red-green, and blue-yellow. It is based on the ratio of the response amount of the receiver, and it explains the empirical fact that there is no yellowish red but yellowish red, and well represents human color vision characteristics . w / k is a luminance component, r / g is a chromaticity component for red-green, and b / y is a chromaticity component for blue-yellow.

  The conversion from the tristimulus values (X, Y, Z) to the values (wk, rg, by) of the opposite color components (w / k, r / g, b / y) is as follows (formula (10)) Is performed by the matrix operation shown in FIG. FIG. 8 shows the spectral characteristics of the color components opposite to the tristimulus values.

  9 and 10 are flowcharts showing the operation procedure of the image quality inspection apparatus 2A (FIG. 7).

  First, in the “image correction process”, in step S501, as a pre-process, shading correction and reduction processes are performed on each input RGB image in order to correct optical characteristics, deal with defect sizes, remove high-frequency noise, and the like. And smoothing processing.

  Next, in the “opposite color image generation process”, in step S502, the RGB image is converted into three tristimulus value images (X, Y, Z) and output. Here, when an XYZ image is directly input, the process of step S502 is not performed. Next, in step S503, the input three tristimulus value images are converted into three opposite color component images (w / k, r / g, b / y) and output.

  In the “area setting process”, in step S504, the size, movement width, and the like of the calculation area to be subjected to the “feature amount calculation process” and “determination process” are set.

  In “feature amount calculation processing”, processing is performed on the calculation divided regions in the respective opposite color component images. In the “feature amount calculation process”, in step S505, a contrast image of the calculation area with each opposite color component set in the “area setting process” is generated.

  In step S506, the contrast image is corrected by a window function in order to cope with the signal discontinuity at the time of FFT.

In step S507, the corrected contrast image is subjected to processing such as two-dimensional Fourier transform, and frequency components Power _wk (u _h , u _v ) and Power _rg (u _h , u _v ) that form a contrast image in each opposite color component. ), Power _by (u _h , u _v ) is extracted.

Next, in step S508, in order to extract dominant frequency components in the calculation region, FFT images Power _wk (u _h , u _v ), Power _rg (u _h , u _v ), Power _by for each opposite color component are extracted. A binarization process is performed for (u _h , u _v ).

  Next, in step S509, the dominant frequency component in the calculation region is extracted by multiplying the FFT binary image and the FFT image in each opposite color component.

On the other hand, in step S510, the respective opposite colors using the spatial frequencies u _h and u _v , the background values wk _BG , rg _BG and by _{BG of} the opposite color components, the sizes X and Y of the defect candidate regions, the angle direction φ, and the like as parameters, respectively. Two-dimensional MTF characteristics S _wk (u _h , u _v , wk _BG , X, Y, φ), S _rg (u _h , u _v , rg _BG ) in the component (w / k, r / g, b / y) X, Y, φ) and S _by (u _h , u _v , by _BG , X, Y, φ) are calculated.

Next, in step S511, the obtained two-dimensional MTF characteristics S _wk (u _h , u _v , wk _BG , X, Y, φ), S _rg (u _h , u _v , rg _BG , X, Y, φ) ), S _by (u _h , u _v , by _BG , X, Y, φ) and frequency components Power _wk (u _h , u _v ), Power _rg (u _h , u _v ), Power _by (u _h , u _{From v} ), the defect intensity (Sem) in each opposite color component (w / k, r / g, b / y) is calculated.

In the “determination process”, in step S512, the defect strengths “Sem _wk ”, “Sem _rg ” in the opposite color components (w / k, r / g, b / y) calculated by the “feature amount calculation process”, _By integrating “Semu _by ”, the total defect intensity (Semu) as color unevenness is calculated.

  In step S513, it is determined whether the defect is a defect by comparing the calculated total defect intensity (Semu) with a preset threshold value. If it is determined that the defect is a defect, step S513 is performed. In S514, information on the detected defects is collected into output data.

  In this way, by calculating the defect intensity in the opposite color component, even for color unevenness, an area that is noticeable by humans is extracted as a defect area without being caught by the defect shape and size, and the user can intuitively It is possible to quantify the defect strength as an objective measurement that is easy to understand.

  In addition, since the defect strength is determined using the frequency components of the entire divided area, not only single defects that do not have a specific frequency component but also periodic defects are not caught by the frequency and direction. , Detection becomes possible.

  In addition, it is possible to detect a new defect because it is determined whether or not it is a defect according to the conspicuousness of each divided area without being aware of the defect shape, reducing the development cost of a new inspection algorithm, and the inspection algorithm development period. Can be shortened. In addition, since there is no extraction process that depends on the defect shape and size, it is possible to achieve a significant reduction in the number of tuning steps related to defect area extraction.

  As shown in FIG. 11, the camera output value may be compensated using the camera output value / tristimulus value conversion parameter stored in the storage device 27. Here, the camera output value / tristimulus value conversion parameter is passed to the opposite color image generation means 26. As described above, the camera output value is compensated using the camera output value / tristimulus value conversion parameter, and the information on the actual tristimulus value is input, so that the highly accurate color unevenness inspection can be performed.

  FIG. 12 is a block diagram showing a configuration for calculating the defect intensity using the recognition limit sensitivity model parameter. In this example, as shown in FIG. 12, the recognition limit sensitivity model parameter is stored as an external file 28, and the recognition limit sensitivity model parameter is input from the external file 28 to the defect strength calculation means 22 and the determination means 23. Here, the recognition limit sensitivity model parameter is passed to some or all of defect candidate extraction processing, feature amount calculation processing, and determination processing corresponding to each opposite component.

  Thus, by inputting the recognition limit sensitivity model parameter from the outside, the maintainability is improved and the sensitivity calculation with high accuracy becomes possible. For example, by measuring the characteristics of a device to be inspected in real time and generating and using a recognition limit sensitivity model parameter from the data, it is possible to calculate the defect intensity corresponding to variations in characteristics between devices and changes over time.

  FIG. 13 is a block diagram illustrating an example in which the image quality inspection method of the present invention is applied to an image quality inspection of an imaging device. As shown in FIG. 13, in this case, an image obtained by imaging the reference light source 30 by the 3CCD camera 10 that is an imaging device to be inspected is input to the image inspection apparatus 2. In the example of FIG. 13, three RGB color images are input for color, but when the inspection target is monochrome, one monochrome image is input.

  This inspection method is applicable to CCD, CMOS, and other imaging devices. Also applicable to monochrome and color imaging devices

  FIG. 14 is a block diagram showing an example in which the image quality inspection method of the present invention is applied to the image quality inspection of the projection device 40. In this example, an image 41 projected from the projection device 40 to be inspected is imaged by being decomposed into RGB by the CCD camera 1 combined with the turret 11. The captured image is input to the image quality inspection apparatus 2A.

  In the example of FIG. 14, three RGB color images are input for color, but when the inspection target is monochrome, one monochrome image is input. This image quality inspection method can be applied to projector liquid crystal and other projection devices. It can also be applied to both monochrome and color projection devices.

  15 and 16 are block diagrams showing a configuration example in which the image quality inspection method of the present invention is applied to an image difference evaluation apparatus.

  As shown in FIG. 15, image data B to be compared with image data A (for example, an image compressed from image data A or an image transmitted from image data A) is input to image difference evaluation device 4, and image data A is The correction means 41a, the defect strength calculation means 42a and the determination means 43a quantify the defect strength of the image data B by the correction means 41b, the defect strength calculation means 42b and the determination means 43b. Furthermore, the defect strength of both is compared by the comparison / determination unit 45 to inspect the change in the image.

  The configuration shown in FIG. 16 corresponds to the RGB color image, and the image data B to be compared with the image data A is input to the image difference evaluation device 5, and the correction means 51a and the opposite color generation means are applied to the image data A. 52a, the defect intensity calculating means 53a, and the determining means 54a quantify the defect intensity of the image data B by the correcting means 51b, the opposite color generating means 52b, the defect intensity calculating means 53b, and the determining means 54b. Furthermore, the defect strength of both is compared by the comparison / determination unit 55 to inspect the change of the image.

As described above, by applying the image quality inspection method of the present invention to the image difference evaluation apparatus, the difference / deterioration of the image can be evaluated with a quantitative value based on the visual recognition limit sensitivity, and the evaluation that is easy to understand objectively. The value can be calculated. Moreover, the deterioration with respect to each shape can be evaluated. For example, it can be used for the following objects.
(1) Evaluation of image compression performance This is performed by setting the images before and after compression as image data A and B.
(2) Image transmission performance evaluation This is performed by setting the images before and after transmission to the image data A and B.
(3) An inspection image of a printed matter change is performed by setting an image obtained by imaging a printed matter as image data A, and then setting an image obtained by capturing the printed matter as image B.
(4) Inspection data for changes in painted object, stability of object position accuracy, or addition of illumination condition correction function and object position correction function by image processing, and image data of image of painted object as master image data This is performed by setting an image obtained by capturing an image of what has been painted to A in image B.

  As described above, according to the image quality inspection apparatus of the present invention, since the influence of discontinuity at the boundary of the inspection area is suppressed using the window function, the image quality is appropriately evaluated without being affected by the inspection area. it can. Further, according to the image quality inspection apparatus of the present invention, the defect intensity is calculated by integrating the frequency characteristic of the visual sensitivity and the frequency component of the two-dimensional image to be inspected. The defect strength can be obtained as a typical measurement value.

  The scope of application of the present invention is not limited to the above embodiment. The present invention can be widely applied to an image quality inspection apparatus and an image quality inspection method for inspecting the image quality of a two-dimensional image.

1 is a block diagram illustrating a configuration of an image quality inspection apparatus according to an embodiment. The flowchart which shows the test | inspection procedure by the image quality inspection apparatus using the MTF characteristic with respect to a brightness | luminance. It is a figure which shows the behavior of window function correction | amendment, (a) is an original image, (b) is a figure which shows the correction image by a window function, respectively. The figure which shows the image quality inspection method which test | inspects, shifting the inspection area | region which extracts a defect sequentially so that it may overlap. The flowchart which shows the test | inspection procedure of FIG. The flowchart which shows the test | inspection procedure at the time of adding a multistep smooth reduction function. The block diagram which shows the structure of the image quality inspection apparatus which test | inspects a color nonuniformity. The figure which shows the spectral characteristic of a color component opposite to a tristimulus value. The flowchart which shows the operation | movement procedure of the image quality inspection apparatus of FIG. The flowchart which shows the operation | movement procedure of the image quality inspection apparatus of FIG. The block diagram which shows the structure which compensates a camera output value using a camera output value / tristimulus value conversion parameter. The block diagram which shows the structure which calculates defect intensity | strength using a recognition limit sensitivity model parameter. 1 is a block diagram illustrating an example in which an image quality inspection method of the present invention is applied to an image quality inspection of an imaging device. The block diagram which shows the example which applied the image quality inspection method of this invention to the image quality inspection of a projection device. The block diagram which shows the structural example which applied the image quality inspection method of this invention to the image difference evaluation apparatus. The block diagram which shows the structural example which applied the image quality inspection method of this invention to the image difference evaluation apparatus. The flowchart which shows the test | inspection procedure by the image quality inspection apparatus using the MTF characteristic with respect to a brightness | luminance. It is a figure which shows a visual sensitivity characteristic etc., (a) is a figure which shows a light and dark sine wave-like stripe pattern, (b) is a figure which shows the human visual sensitivity characteristic with respect to a spatial frequency, (c) is a range of a certain visual angle. The figure which shows the state which displayed the sine wave of a certain wave number.

Explanation of symbols

2 Image quality inspection device 2A Image quality inspection device 4 Image difference evaluation device (image quality inspection device)
5 Image difference evaluation device (image quality inspection device)
21 Correction means 22 Defect strength calculation means 23 Determination means 43a Determination means 43b Determination means 45 Determination comparison unit (determination means)
54a determination means 54b determination means 55 determination comparison unit (determination means)

Claims (6)

In an image quality inspection device that inspects the image quality of a two-dimensional image,
Correction means for performing correction so as to suppress the influence of discontinuity at the boundary of the inspection region using a window function for the inspection region of the two-dimensional image to be inspected,
Determination means for determining the image quality of the two-dimensional image based on the image after correction by the correction means;
Control means for controlling to repeat the correction by the correction means and the determination by the determination means while sequentially switching the inspection areas obtained by dividing the two-dimensional image into a plurality of areas overlapping each other;
Defect strength calculating means for calculating defect strength by integrating the frequency characteristics of visual sensitivity and the frequency component of the image after being corrected by the correcting means;
With
The two-dimensional image Ri inverse colored image der,
Before Symbol judging means judges the quality of the two-dimensional image on the basis of the defect intensity calculated by the defect intensity calculating means,
Before SL defect intensity calculating means, image quality inspection system you and calculates the defect intensity only the dominant frequency component in the two-dimensional frequency region of the examination region. The image quality inspection apparatus according to claim 1, wherein the dominant frequency component is extracted by multiplying a Fourier transform image of the inspection region and a binarized image of the Fourier transform image . The defect intensity calculating means calculates the first defect intensity by integrating the frequency characteristic of the visual sensitivity and the frequency component of the first two-dimensional image to be inspected, and the frequency characteristic of the visual sensitivity. And calculating the second defect strength by integrating the frequency components of the second two-dimensional image to be inspected,
The determination unit compares the first defect intensity and the second defect intensity calculated by the defect intensity calculation unit to compare an image difference between the first two-dimensional image and the second two-dimensional image. The image quality inspection device according to claim 1 , wherein the image quality inspection device is determined. The two-dimensional image quality inspection apparatus according to any one of claims 1 to 3, characterized in that a luminance image. Quality inspection apparatus according to any one of claims 1 to 4, characterized in that the size of the inspection area is variable. In an image quality inspection method for inspecting the image quality of a two-dimensional image,
A step of performing correction so as to suppress the influence of discontinuity at the boundary of the inspection region by using a window function for the inspection region of the two-dimensional image to be inspected;
Determining the image quality of the two-dimensional image based on the image corrected by the correcting step;
Controlling to repeat the step of performing the correction and the step of determining while sequentially switching the inspection regions obtained by dividing the two-dimensional image into a plurality of regions overlapping each other;
Calculating the defect strength by integrating the frequency characteristics of the visual sensitivity and the frequency components of the image after being corrected by the correcting step;
With
The two-dimensional image Ri inverse colored image der,
In the step of performing a pre-Symbol decision, it determines the image quality of the two-dimensional image on the basis of the defect intensity calculated by the step of calculating the defect intensity,
In the step of calculating a pre-Symbol defect strength, image quality inspection how to and calculating a defect intensity of only the dominant frequency component in the two-dimensional frequency region of the examination region.